22. Introduction to Agentic AI#
Agentic AI tools use a large language model to plan and carry out multi-step tasks, such as writing and running code, rather than just answer a question. The Agentic AI section has five chapters:
Introduction to Agentic AI (this chapter) defines the Key Concepts the rest of the section uses and surveys the current landscape in Agentic AI Tools: coding agents, research and science agents, and the frameworks and protocols behind them.
Agents on the Cluster covers setting up and running these tools on the cluster, a first agentic task, agent security, and serving open-weight models on the cluster’s GPUs.
Configuring and Extending Agents covers giving an agent your project’s conventions, workflows, and guardrails, and connecting it to your own tools and data.
Agentic Workflows in Practice puts agents to work on research tasks: SLURM jobs, experiment management, unfamiliar codebases, refactoring, literature review and data exploration, and multi-agent workflows.
Trustworthy Agentic Research covers evaluating and monitoring what an agent does, and fitting agents into a research workflow while keeping their output trustworthy.
New to agents? Start with Key Concepts, then Using Agentic AI on the Cluster and Your First Agentic Workflow on the Cluster, and read Agent Security and Scoping before you give an agent real work. For the data-handling rules that govern all of this, see Security and Compliance.